I85PREP
⌘ K
AAI-M study materials
AAI-M study resources

AAI-M practice questions

One question at a time. Build your AAI-M understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1AI Fundamentals and Terminology

Which term describes a machine learning model that has been trained on data and is now being used to make predictions on new, unseen inputs?

  1. Validation
  2. Regularization
  3. Training
  4. Inference Correct answer.

Explanation

Inference is the process of using a trained model to generate predictions or outputs for new data. Training is the earlier phase where the model learns from data, while validation and regularization are techniques used during development.

Question 2AI Fundamentals and Terminology

In supervised learning, what is the term for the known output that accompanies each input example during training?

  1. Hyperparameter
  2. Weight
  3. Feature
  4. Label Correct answer.

Explanation

In supervised learning, each training example consists of input features and a corresponding label (or target) that the model learns to predict. Features are the input variables, hyperparameters are configuration settings, and weights are learned model parameters.

Question 3AI Fundamentals and Terminology

Which type of machine learning is used when the goal is to discover natural groupings in data without any predefined categories?

  1. Reinforcement learning
  2. Semi-supervised learning
  3. Unsupervised learning Correct answer.
  4. Supervised learning

Explanation

Unsupervised learning works with unlabeled data to find patterns such as clusters or associations. Supervised learning requires labeled examples, reinforcement learning uses reward signals, and semi-supervised learning combines labeled and unlabeled data.

Question 4AI Fundamentals and Terminology

What does the term 'overfitting' describe in machine learning?

  1. A model that has been trained on too little data
  2. A model that uses too many regularization techniques
  3. A model that performs well on training data but poorly on new data Correct answer.
  4. A model that is too simple to capture the underlying pattern

Explanation

Overfitting occurs when a model learns the training data too closely, including its noise, and therefore generalizes poorly to unseen data. A model that is too simple exhibits underfitting, not overfitting.

Question 5AI Fundamentals and Terminology

Which of the following best describes a neural network?

  1. A rule-based system that follows hand-coded logic
  2. A statistical model that stores all training examples for lookup
  3. A model composed of layers of interconnected units that transform inputs Correct answer.
  4. A decision tree that splits data on a single feature at a time

Explanation

A neural network is built from layers of interconnected units (neurons) that apply learned weights and nonlinear activations to transform inputs into outputs. It is not a rule-based system, a nearest-neighbor lookup store, or a single-feature decision tree.

Question 6AI Fundamentals and Terminology

What is the primary purpose of a loss function in model training?

  1. To split the dataset into training and test sets
  2. To initialize the model's weights randomly
  3. To select the learning rate for the optimizer
  4. To measure how far the model's predictions are from the target values Correct answer.

Explanation

A loss function quantifies the discrepancy between predictions and target values, providing the signal that optimization minimizes. Data splitting, weight initialization, and learning-rate selection are separate procedures.

Question 7AI Fundamentals and Terminology

In the context of large language models, what does 'token' refer to?

  1. A single floating-point number in the model's output
  2. A unique identifier for each user session
  3. A security credential used for API access
  4. A unit of text such as a word or subword piece Correct answer.

Explanation

Language models process text by breaking it into tokens, which are typically words or subword pieces. Tokens are the basic units the model reads and generates, not session IDs, credentials, or individual output numbers.

Question 8AI Fundamentals and Terminology

Which term refers to the ability of a trained model to perform well on data it has not seen before?

  1. Generalization Correct answer.
  2. Capacity
  3. Convergence
  4. Regularization

Explanation

Generalization is the model's ability to make accurate predictions on new, unseen data. Convergence refers to training stabilizing, capacity is the model's representational power, and regularization is a technique to improve generalization.

Question 9AI Fundamentals and Terminology

A team trains a model to classify emails as spam or not spam using a dataset where each email is labeled. Which learning paradigm does this represent?

  1. Unsupervised learning
  2. Reinforcement learning
  3. Self-supervised learning
  4. Supervised learning Correct answer.

Explanation

Because each training email has a known label (spam or not spam), the task is supervised learning. Unsupervised learning lacks labels, reinforcement learning uses reward signals, and self-supervised learning generates labels from the data itself rather than using provided labels.

Question 10AI Fundamentals and Terminology

A model achieves 99% accuracy on its training set but only 60% on a held-out test set. Which explanation is most consistent with this result?

  1. The model is underfitting the training data
  2. The test set is too small to be meaningful
  3. The model has too few parameters
  4. The model is overfitting the training data Correct answer.

Explanation

A large gap between high training accuracy and much lower test accuracy is the classic signature of overfitting, where the model has memorized training-specific patterns. Underfitting would show poor performance on both sets, and too few parameters would typically reduce training accuracy as well.

Question 11AI Fundamentals and Terminology

Which statement best distinguishes a parameter from a hyperparameter in a machine learning model?

  1. Parameters are only used at inference time; hyperparameters are only used during training
  2. Parameters and hyperparameters are both learned during training
  3. Parameters are learned from data during training; hyperparameters are set before training Correct answer.
  4. Parameters are set by the user before training; hyperparameters are learned from data

Explanation

Model parameters, such as weights, are learned from the training data, while hyperparameters, such as learning rate or number of layers, are configured before training. The other options reverse or misstate these roles.

Question 12AI Fundamentals and Terminology

What is the main role of an activation function in a neural network?

  1. To introduce nonlinearity so the network can learn complex patterns Correct answer.
  2. To normalize the input features before training
  3. To compute the loss between predictions and targets
  4. To update the weights using gradients

Explanation

Activation functions apply a nonlinear transformation to neuron outputs, enabling the network to model complex, nonlinear relationships. Normalization, loss computation, and weight updates are handled by other components.

Question 13AI Fundamentals and Terminology

A data scientist wants to reduce the number of input features while preserving as much variance as possible. Which technique is designed for this purpose?

  1. k-means clustering
  2. Gradient boosting
  3. Principal component analysis Correct answer.
  4. Logistic regression

Explanation

Principal component analysis (PCA) is a dimensionality-reduction technique that projects data onto directions of maximum variance. Logistic regression and gradient boosting are predictive models, and k-means is a clustering algorithm.

Question 14AI Fundamentals and Terminology

Which of the following is an example of a classification task rather than a regression task?

  1. Forecasting tomorrow's temperature
  2. Predicting whether a customer will churn Correct answer.
  3. Predicting the selling price of a house
  4. Estimating the number of units sold next month

Explanation

Classification predicts a discrete category, such as whether a customer will churn or not. The other options predict continuous numeric values, which are regression tasks.

Question 15AI Fundamentals and Terminology

What does the term 'batch size' refer to in training a neural network?

  1. The number of epochs used for training
  2. The total number of layers in the network
  3. The number of training examples processed before one parameter update Correct answer.
  4. The fraction of data reserved for testing

Explanation

Batch size is the number of training examples used to compute one gradient update. Epochs count full passes over the data, layers are architectural components, and the test fraction relates to data splitting.

Question 16AI Fundamentals and Terminology

Which statement best describes the difference between artificial intelligence, machine learning, and deep learning?

  1. Machine learning is a subset of deep learning, which is a subset of AI
  2. They are three names for the same field
  3. Deep learning is a subset of machine learning, which is a subset of AI Correct answer.
  4. Artificial intelligence is a subset of machine learning, which is a subset of deep learning

Explanation

Artificial intelligence is the broad goal of building intelligent systems; machine learning is a subset of AI focused on learning from data; deep learning is a subset of machine learning using multi-layer neural networks. The other options invert or misstate this hierarchy.

Question 17AI Fundamentals and Terminology

A model trained on historical loan data consistently approves loans for one demographic group and rejects another, even when financial profiles are similar. Which concept does this illustrate?

  1. Concept drift
  2. Algorithmic bias Correct answer.
  3. Overfitting
  4. Data leakage

Explanation

Algorithmic bias occurs when a model systematically produces unfair outcomes for certain groups, often reflecting biases in the training data. Data leakage, overfitting, and concept drift describe different problems related to data handling and model performance.

Question 18AI Fundamentals and Terminology

What is the purpose of a validation set during model development?

  1. To increase the size of the training data
  2. To tune hyperparameters and compare model variants Correct answer.
  3. To train the model's parameters
  4. To provide the final unbiased estimate of model performance

Explanation

A validation set is used to tune hyperparameters and select among model variants without touching the test set. The training set fits parameters, and the test set provides the final performance estimate.

Question 19AI Fundamentals and Terminology

In reinforcement learning, what is the role of the reward signal?

  1. It defines the environment's state transition probabilities
  2. It provides feedback that the agent seeks to maximize over time Correct answer.
  3. It specifies the set of actions available to the agent
  4. It initializes the agent's policy network

Explanation

The reward signal tells the agent how good an action or state is, and the agent learns a policy that maximizes cumulative reward. State transitions, action sets, and initialization are separate elements of the reinforcement learning setup.

Question 20AI Fundamentals and Terminology

Which of the following best characterizes a foundation model?

  1. A model trained on broad data that can be adapted to many downstream tasks Correct answer.
  2. A small model trained for one narrow task
  3. A model that only performs arithmetic operations
  4. A model that requires no training data

Explanation

Foundation models are trained on large, broad datasets and can be adapted (for example, via fine-tuning or prompting) to a wide range of downstream tasks. They are not task-specific small models, arithmetic-only systems, or models that need no data.

Question 21AI Fundamentals and Terminology

A model performs well in testing but degrades over time in production as user behavior changes. Which term best describes this phenomenon?

  1. Data leakage
  2. Concept drift Correct answer.
  3. Underfitting
  4. Overfitting

Explanation

Concept drift occurs when the statistical relationship between inputs and outputs changes over time, causing a previously accurate model to degrade. Overfitting and underfitting relate to model fit, and data leakage is a training-set contamination issue.

Question 22AI Fundamentals and Terminology

Which statement accurately describes the relationship between a model's training loss and its validation loss during typical overfitting?

  1. Validation loss decreases while training loss increases
  2. Both losses remain constant after the first epoch
  3. Training loss decreases while validation loss eventually begins to increase Correct answer.
  4. Both training and validation loss decrease indefinitely

Explanation

During overfitting, training loss continues to fall as the model memorizes training data, but validation loss reaches a minimum and then rises, indicating worsening generalization. The other patterns do not match standard overfitting behavior.

Question 23AI Fundamentals and Terminology

A team notices that a model's performance is excellent on a random train/test split but drops sharply when evaluated on data collected from a later time period. Which issue is most likely?

  1. The learning rate was too high
  2. The test set was too large
  3. The random split leaked future information into training Correct answer.
  4. The model has too few layers

Explanation

A random split on time-ordered data can place future examples in the training set, allowing the model to learn from information it would not have at prediction time. This is a form of data leakage that inflates random-split performance relative to a temporal split.

Question 24AI Fundamentals and Terminology

Which of the following best explains why regularization can improve a model's test performance even when it increases training error?

  1. It removes the need for a validation set
  2. It increases the number of parameters the model can learn
  3. It guarantees the global minimum of the loss function
  4. It reduces model capacity to memorize noise, improving generalization Correct answer.

Explanation

Regularization penalizes complexity, which constrains the model's ability to fit noise in the training data and thereby improves generalization to unseen data. It does not add parameters, guarantee a global minimum, or eliminate the need for validation.

Question 25AI Fundamentals and Terminology

A language model is described as 'autoregressive.' What does this mean about how it generates text?

  1. It predicts each next token conditioned on the previously generated tokens Correct answer.
  2. It retrieves complete sentences from a stored database
  3. It generates all tokens simultaneously in one step
  4. It generates tokens in a random order until a stop signal

Explanation

An autoregressive model generates text one token at a time, conditioning each prediction on the tokens produced so far. It does not emit all tokens at once, retrieve stored sentences, or generate in random order.

Question 26AI Fundamentals and Terminology

Which term describes a machine learning system that improves its performance on a task by receiving feedback in the form of rewards and penalties for its actions?

  1. Reinforcement learning Correct answer.
  2. Unsupervised learning
  3. Supervised learning
  4. Semi-supervised learning

Explanation

Reinforcement learning trains an agent through trial and error, using rewards and penalties to shape behavior over time. The other paradigms rely on labeled targets or on finding structure in unlabeled data rather than on evaluative feedback from an environment.

Question 27AI Fundamentals and Terminology

In natural language processing, what does the term 'embedding' most commonly refer to?

  1. A rule-based grammar that parses sentences into syntax trees
  2. A dense numeric vector that represents a word or other item in a continuous space Correct answer.
  3. A lookup table that stores the raw character count of each document
  4. A compression scheme that shortens text before it is stored

Explanation

An embedding maps discrete items such as words into dense vectors so that items with similar meaning end up close together in the vector space. These vectors are learned from data and used as numeric inputs to downstream models.

Question 28AI Fundamentals and Terminology

A system is described as using 'inference' rather than 'training.' Which activity is taking place?

  1. Hyperparameters are tuned by comparing validation scores
  2. The model's weights are adjusted using backpropagation
  3. A new dataset is collected and labeled for future use
  4. The trained model is applied to new inputs to produce predictions Correct answer.

Explanation

Inference is the forward pass in which an already-trained model processes new inputs and returns outputs. Weight updates and hyperparameter tuning belong to the training and model-selection phases that precede deployment.

Question 29AI Fundamentals and Terminology

Which statement best describes what is meant by 'human-in-the-loop' in an AI system?

  1. People review, correct, or approve model outputs as part of the operating workflow Correct answer.
  2. Humans manually label every record in the training set before any model is built
  3. The system runs only when a person is physically present at the server
  4. The model is retrained entirely by hand without any automated optimization

Explanation

Human-in-the-loop describes an operational design in which people remain part of the process, checking or correcting model outputs and feeding that feedback back into the system. It is about ongoing oversight and intervention, not about how the model is trained or hosted.

Question 30AI Fundamentals and Terminology

A vendor claims its model is 'multimodal.' Which capability does this claim most directly imply?

  1. The model was trained by more than one organization jointly
  2. The model accepts and relates more than one type of input, such as text and images Correct answer.
  3. The model supports several programming language APIs for access
  4. The model can be deployed in several cloud regions at once

Explanation

Modality refers to the type of data, such as text, images, or audio, so a multimodal model processes and relates multiple data types within one system. Deployment locations, API languages, and the number of training organizations are unrelated to the term.

Question 31Data Literacy and Data Quality

A dataset column stores customer ages, but one record contains the value -12. Which data quality dimension is most directly violated?

  1. Validity Correct answer.
  2. Consistency
  3. Timeliness
  4. Uniqueness

Explanation

Validity concerns whether values conform to defined rules or allowable ranges for their domain. An age of -12 falls outside the permitted range, so it is invalid regardless of when it was captured or whether it is duplicated.

Question 32Data Literacy and Data Quality

Two systems record the same customer's shipping address, and the values differ. Which data quality dimension does this most directly illustrate a failure of?

  1. Accessibility
  2. Consistency Correct answer.
  3. Completeness
  4. Precision

Explanation

Consistency measures whether the same fact is represented the same way across records, systems, or time. Conflicting addresses for one entity signal a consistency problem across sources.

Question 33Data Literacy and Data Quality

A required field is left blank in 30 percent of records in a customer table. Which data quality dimension is primarily affected?

  1. Accuracy
  2. Timeliness
  3. Completeness Correct answer.
  4. Validity

Explanation

Completeness describes the degree to which required data is present. Blank values in a mandatory field are missing data, which is the defining concern of completeness.

Question 34Data Literacy and Data Quality

Which statement best describes the relationship between data quality and data governance?

  1. Quality is purely a technical concern handled by database administrators
  2. Governance sets decision rights and standards that quality management then enforces Correct answer.
  3. Quality management replaces governance once automated checks are deployed
  4. Governance applies only to financial reporting data

Explanation

Data governance establishes accountability, policies, and standards, while quality management operationalizes those standards through measurement and remediation. The two are complementary rather than substitutes.

Question 35Data Literacy and Data Quality

A report shows a metric of 12.0 where the underlying source value was 1.20. Which type of data quality issue is this?

  1. A completeness defect
  2. A timeliness defect
  3. A referential integrity defect
  4. A transformation or scaling error Correct answer.

Explanation

A value that is off by a consistent factor of ten typically results from a unit conversion or scaling mistake during transformation or loading. The record is present and current, so the defect is in how the value was processed.

Question 36Data Literacy and Data Quality

What is the primary purpose of defining a data quality rule such as 'order_date must be less than or equal to ship_date'?

  1. To encrypt sensitive date fields at rest
  2. To reduce the storage footprint of the warehouse
  3. To automate detection of records that violate an expected relationship Correct answer.
  4. To document the business meaning of the table for new analysts

Explanation

Quality rules encode expected constraints so that violations can be detected systematically and repeatedly. They translate business expectations into testable conditions.

Question 37Data Literacy and Data Quality

Which practice most directly supports reproducibility of a data quality assessment?

  1. Re-running checks ad hoc whenever a stakeholder asks
  2. Recording the rule definitions, source snapshot, and run date used Correct answer.
  3. Relying on the analyst's memory of prior findings
  4. Reporting only the pass or fail outcome without counts

Explanation

Reproducibility requires that another person can repeat the assessment and obtain the same result. Capturing rule logic, the data snapshot, and the execution date makes the result auditable and repeatable.

Question 38Data Literacy and Data Quality

A data steward finds that a customer email field contains values with no '@' symbol. Which quality check category best describes the rule being applied?

  1. Duplicate detection
  2. Referential integrity checking
  3. Volume reconciliation
  4. Format or pattern validation Correct answer.

Explanation

Format validation tests whether values match an expected syntactic pattern, such as an email address containing an '@' and a domain. It does not test existence, uniqueness, or row counts.

Question 39Data Literacy and Data Quality

An analyst computes a completeness score as the number of non-null values divided by the total number of records for a required field. If 1,850 of 2,000 records are populated, what is the completeness score?

  1. 0.850
  2. 0.925 Correct answer.
  3. 0.075
  4. 0.150

Explanation

Completeness is populated records divided by total records: 1,850 divided by 2,000 equals 0.925. The complement, 0.075, is the null rate.

Question 40Data Literacy and Data Quality

A retail firm discovers that the same customer exists three times under slightly different name spellings. Which quality dimension is most directly implicated, and what is the typical remedy?

  1. Validity; enforce a domain list on the name field
  2. Accuracy; recapture the values from source documents
  3. Timeliness; refresh the dataset more frequently
  4. Uniqueness; apply entity resolution and deduplication Correct answer.

Explanation

Duplicate representations of one real-world entity are a uniqueness problem, often addressed through matching and merging in entity resolution. Recapturing or reformatting values does not by itself resolve the duplication.

Question 41Data Literacy and Data Quality

Which metric best captures the timeliness of a daily data feed against a service-level objective of availability by 06:00?

  1. Percentage of non-null values in the feed
  2. Count of distinct source systems feeding the table
  3. Number of columns in the feed
  4. Percentage of days the feed was available by 06:00 Correct answer.

Explanation

Timeliness against an availability objective is measured by the proportion of periods in which the data was available by the stated deadline. The other measures describe structure or completeness rather than currency.

Question 42Data Literacy and Data Quality

A data quality dashboard shows a validity pass rate of 99.8 percent, yet downstream users report frequent errors. What is the most likely explanation?

  1. Validity rules are always unreliable
  2. The rules cover only a narrow set of columns or conditions Correct answer.
  3. Users are misinterpreting correct data
  4. The dashboard is calculating the rate incorrectly in all cases

Explanation

A high pass rate on a limited rule set can coexist with real defects in areas the rules do not cover. Coverage of rules across critical data elements must be assessed alongside pass rates.

Question 43Data Literacy and Data Quality

In root cause analysis of a recurring data quality defect, which step should come immediately after confirming and documenting the defect?

  1. Adding a new dashboard
  2. Deploying a permanent code fix
  3. Quantifying impact and tracing the defect to its origin point in the data flow Correct answer.
  4. Closing the incident ticket

Explanation

Effective remediation begins with understanding scope and origin so the fix addresses the cause rather than a symptom. Tracing to the point of introduction in the pipeline informs both impact assessment and remedy selection.

Question 44Data Literacy and Data Quality

A referential integrity check on an orders table finds 42 order rows whose customer_id has no matching row in the customers table. What does this finding indicate?

  1. The customer table is over-normalized
  2. The order amounts are inaccurate
  3. Orphaned foreign key values exist in the orders table Correct answer.
  4. The orders table contains duplicate records

Explanation

Referential integrity requires that foreign key values resolve to an existing parent record. Unmatched keys are orphaned references, which may stem from late-arriving or deleted parent records.

Question 45Data Literacy and Data Quality

Which approach best establishes a baseline for measuring improvement in data quality over time?

  1. Measure the same defined rules on the same scope at regular intervals Correct answer.
  2. Change the rule set each quarter to reflect new priorities
  3. Survey users once at project kickoff
  4. Compare current results only to vendor benchmarks

Explanation

Trend measurement requires a stable definition and scope so that changes in the metric reflect changes in the data rather than changes in the test. Consistent periodic measurement provides a comparable baseline.

Question 46Data Literacy and Data Quality

A field labeled 'revenue' is populated with values that include tax, while the finance team's definition excludes tax. Which data quality concern does this raise?

  1. A referential integrity failure
  2. A timeliness defect in the revenue feed
  3. A duplicate record problem
  4. A metadata and definition mismatch affecting accuracy of interpretation Correct answer.

Explanation

When a stored value does not match the documented business definition, users draw incorrect conclusions even if the number is internally consistent. Clear, agreed definitions in metadata are essential to interpret accuracy.

Question 47Data Literacy and Data Quality

Which control most directly prevents invalid values from entering a system at the point of capture?

  1. Annual data audits
  2. Input validation with allowable value lists and range checks Correct answer.
  3. Periodic reconciliation reports
  4. Archiving historical records

Explanation

Preventive controls at entry stop defects before they propagate. Validation rules at capture reduce downstream remediation effort compared with detection-only controls.

Question 48Data Literacy and Data Quality

A team is prioritizing remediation across five data elements. Which combination of factors best supports prioritization?

  1. Alphabetical order of field names and team size
  2. Number of columns in the table and storage cost
  3. Age of the source system and vendor contract terms
  4. Business criticality of the element and severity and frequency of the defect Correct answer.

Explanation

Prioritization should reflect the business impact of the data element together with how severe and how frequent its defects are. These factors together estimate risk and guide where effort yields the most value.

Question 49Data Literacy and Data Quality

An organization defines a critical data element and assigns a steward, a definition, quality rules, and a remediation owner. Which governance mechanism does this most closely represent?

  1. A data quality ownership and accountability model Correct answer.
  2. A data retention schedule
  3. A disaster recovery plan
  4. A schema migration strategy

Explanation

Assigning accountable owners, agreed definitions, and rules for critical elements is the core of a quality ownership model. It links accountability to measurable expectations.

Question 50Data Literacy and Data Quality

A pipeline applies a rule that truncates text to 50 characters, silently cutting off part of a customer address. Which quality dimension is most directly harmed, and why?

  1. Timeliness, because the address may be outdated
  2. Uniqueness, because addresses may repeat
  3. Consistency, because addresses vary in format
  4. Completeness, because required information is removed even though the field is populated Correct answer.

Explanation

Truncation removes part of the required content, so the record is populated but not fully complete for its purpose. Completeness is about the presence of all needed content, not merely a non-null value.